Research On Whole Field Intelligent Detection Method Of Steel Bridge Rivet Disease Based On UAV

Due to the harsh environment,fatigue loads,and the impact of high-speed trains,rivets on bridge often suffer from diseases.In actual engineering,rivet disease is mainly through regular inspections of bridge maintenance personnel on the bridge or visual inspection with the help of telescopes.The detection cost is high,efficiency is low,detection results are highly subjective,and there is a potential threat to the personal safety of bridge maintenance personnel.With the rapid development of artificial intelligence,non-destructive detection methods based on computer vision are becoming more and more mature,this topic is based on the UAV,computer vision technology and deep learning algorithm to study the full-field intelligent detection method for rivet diseases of railway steel truss bridges.The main research contents are as follows:(1)Panoramic image stitching technology based on ultra-high resolution.Based on the UAV equipped with high-definition platform camera,the reasonable flight path is planned to obtain the local ultra-high resolution aerial image of the bridge.Aiming at the interference of background redundancy information on image stitching,the foreground extraction methods suitable for ordinary resolution images and ultra-high resolution images are studied respectively.The homography matrix between images is calculated by SIFT feature point matching,and then the two-dimensional image stitching is completed to show the whole field information of bridge structure,which provides the basis for the whole field positioning of rivet diseases.(2)Full-field intelligent recognition method of rivet corrosion based on deep learning.The special dataset of rivet corrosion disease is prepared based on panoramic image.The deep learning model SSD is built for corrosion feature extraction and intelligent recognition.Aiming at the problem of small rivet target and low detection precision,the method of improving model detection precision is studied.Based on the prior information of the space distribution of bridge rivets,the algorithm is optimized to identify and locate the rivet disease types.Through the correlation of the rivet sub-image and the panoramic image,the positioning of the corroded rivet in the bridge is completed.(3)Research on bridge 3D model reconstruction method based on oblique photography.The dataset of 3D model reconstruction is collected based on UAV platform.Aiming at the shadow of bridge backlight caused by sunlight,the local image enhancement method is studied to achieve the enhancement of specific areas of the image.The paper studies the principle of 3D model reconstruction based on oblique photography,and generates three-dimensional dense point cloud based on image pos data and pixel coordinates,so as to realize the 3D real-world model reconstruction of bridges,and more visually display the information of bridge components.